I have a variable reponse with a lot of zeros and positives values in the remaining part and some categorical and quantitive variables as features.
Given a Tweedie or Poisson distribution of Y, why can i apply a Random Forest and/or a Gradient Boosting regressor ? I read that an option is to do a two steps model : first to predict if the Y is zero or non zero, and then to predict the positive values, but i do not understand how we can aggregate the two models. I read too, that is possible to apply a Grandient Boosting for a Tweedie distribution, but i d’ont know how to choose parameters to fit with a tweedie or poisson distribution.
Thank you for yours advises and help !